From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jim Fehlig Subject: Re: That xenstored console leak... Date: Fri, 18 Jan 2008 16:14:16 -0700 Message-ID: <479132C8.2090802@novell.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: References: C3B21B02.122B6%Keir.Fraser@cl.cam.ac.uk List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Keir Fraser Cc: xen-devel@lists.xensource.com, John Levon List-Id: xen-devel@lists.xenproject.org Keir Fraser wrote: >> Reverting changesets 15967 and 15957 in addition to the attached patch >> fixes the leak and allows multiple localhost migrations. I'm not sure >> what we get by nuking /vm//device/vif/ anyway - other >> than the problems we're seeing :-). vif appears to be the only device >> stored in the /vm//device path anyway. >> >> I will continue testing with this setup ... >> > > Isn't having two domains (even from the same vm) pointing at the same /vm/ > path a recipe for further bugs? Most of the lowlevel xend code doesn't seem > to understand the concept that domains can map to the same vm, and could > hence tread on each others toes via the /vm/ path. > > If we revert the two patches, what happens when you create/destroy lots of > domains all with different uuids? I expect the leak will still exist in that > case. > Sorry for the delay. I'm not sure why you think that the leak would still exist with those changesets reverted. 15957 (and subsequently 15967) introduced the leak by creating a whole new /vm/- path, leaving the previous path orphaned. But I certainly don't claim to be an expert on this code so perhaps I'm not understanding your concern. Nevertheless, I created/destroyed lots of domains on 3.2 with those changesets reverted and do not see the leak. However I wouldn't expect so since each domain has a different uuid and hence a different /vm/ path, which is removed when the domain is destroyed. BTW, with those changesets /vm/ path is leaked on save/restore, reboot, and localhost migration. Perhaps the source domain in these operations should be removing its /vm path on destruction? Jim